» 케라스 API / Callbacks API

Callbacks API

A callback is an object that can perform actions at various stages of training (e.g. at the start or end of an epoch, before or after a single batch, etc).

You can use callbacks to:

  • Write TensorBoard logs after every batch of training to monitor your metrics
  • Periodically save your model to disk
  • Do early stopping
  • Get a view on internal states and statistics of a model during training
  • ...and more

Usage of callbacks via the built-in fit() loop

You can pass a list of callbacks (as the keyword argument callbacks) to the .fit() method of a model:

my_callbacks = [
    tf.keras.callbacks.EarlyStopping(patience=2),
    tf.keras.callbacks.ModelCheckpoint(filepath='model.{epoch:02d}-{val_loss:.2f}.h5'),
    tf.keras.callbacks.TensorBoard(log_dir='./logs'),
]
model.fit(dataset, epochs=10, callbacks=my_callbacks)

The relevant methods of the callbacks will then be called at each stage of the training.


Using custom callbacks

Creating new callbacks is a simple and powerful way to customize a training loop. Learn more about creating new callbacks in the guide Writing your own Callbacks, and refer to the documentation for the base Callback class.


Available callbacks